👋 Hi, everyone!
We are ByteDance Seed team.

You can get to know us better through the following channels👇

ByteFF2

This is the source repository for ByteFF-Pol.

• ByteFF-Pol is a polarizable force field parameterized by a graph neural network (GNN), trained on high-level quantum mechanics (QM) data, thus eliminating the need for experimental calibration. ByteFF-Pol achieves exceptional accuracy in predicting the thermodynamic and transport properties of small-molecule liquids and electrolytes, outperforming SOTA traditional and ML force fields.

News

[2025/08/25]🔥We release ByteFF-Pol.

Getting started

Prerequisites

• Python version >= 3.11

Python Dependencies

All required Python packages are listed in requirements.txt. To install them, run:

pip install -r requirements.txt

Installing Gromacs

Download Gromacs from official website.

wget https://ftp.gromacs.org/gromacs/gromacs-2025.3.tar.gz

To install Gromacs, please refer to the official documentation.

tar xfz gromacs-2025.3.tar.gz
cd gromacs-2025.3
mkdir build
cd build
cmake .. -DGMX_BUILD_OWN_FFTW=ON -DREGRESSIONTEST_DOWNLOAD=ON
make
make check
sudo make install
source /usr/local/gromacs/bin/GMXRC

Installing OpenMM for ByteFF2

To run ByteFF2, you need a customized version of OpenMM and OpenMM-VelocityVerlet.

1. Navigate to the submodules/openmm directory:

   cd submodules/openmm

2. Run the installation script:

   ./install.sh [OPENMM_DIR]

   • [OPENMM_DIR] (optional): Installation path for OpenMM.
   • Default installation path is:

     /usr/local/openmm
     

3. The script will:

   • Compile and install the patched openmm (v8.3.1)
   • Compile and install openmm-velocityVerlet
   • Add required environment variables (OPENMM_DIR and LD_LIBRARY_PATH) to your ~/.bashrc

4. After installation, restart your terminal or run:

   source ~/.bashrc

After successful installation, you should see:

Success: Installed OpenMM and openmm-velocityVerlet.

Trained Models

The model and configuration file are available on HuggingFace byteff2.

To download the model and configuration file, run:

pip install -U "huggingface_hub[cli]"
hf download ByteDance-Seed/byteff2 --local-dir byteff2

Quick Start

You can refer to several examples in the · directory; more details are available in the README.md file for each example.

• example/1_training contains scripts for training ByteFF-Pol.
• example/2_compare_qm contains scripts to compare QM and FF energies.
• example/3_write_params contains scripts to generate force field parameters using trained ByteFF-Pol model.
• example/4_MD_simulations contains scripts for molecular dynamics (MD) simulations using ByteFF-Pol.
• example/5_similarity contains scripts for similarity analysis using ByteFF-Pol.

Run Tests

You can verify the environments by running the tests:

make test

License

This project is licensed under the Apache License, Version 2.0.

Citation

If you find ByteFF-Pol or ByteFF is useful for your research and applications, feel free to give us a star ⭐ or cite us using:

@misc{zheng2025bridgingquantummechanicsorganic,
  title         = {Bridging Quantum Mechanics to Organic Liquid Properties via a Universal Force Field},
  author        = {Tianze Zheng and Xingyuan Xu and Zhi Wang and Xu Han and Zhenliang Mu and Ziqing Zhang and Sheng Gong and Kuang Yu and Wen Yan},
  year          = {2025},
  eprint        = {2508.08575},
  archivePrefix = {arXiv},
  primaryClass  = {physics.comp-ph},
  url           = {https://arxiv.org/abs/2508.08575}
}

@Article{D4SC06640E,
  author    = {Tianze Zheng and Ailun Wang and Xu Han and Yu Xia and Xingyuan Xu and Jiawei Zhan and Yu Liu and Yang Chen and Zhi Wang and Xiaojie Wu and Sheng Gong and Wen Yan},
  title     = {Data-driven parametrization of molecular mechanics force fields for expansive chemical space coverage},
  journal   = {Chem. Sci.},
  year      = {2025},
  pages     = {-},
  publisher = {The Royal Society of Chemistry},
  doi       = {10.1039/D4SC06640E},
  url       = {http://dx.doi.org/10.1039/D4SC06640E}
}

About ByteDance Seed Team

Founded in 2023, ByteDance Seed Team is dedicated to crafting the industry's most advanced AI foundation models. The team aspires to become a world-class research team and make significant contributions to the advancement of science and society.